Wikipedia Sucks!

https://arstechnica.com/ai/2025/06/yuck ... or-revolt/
https://mashable.com/article/wikipedia- ... ors-revolt

Reporters only saw it because 404media ran it as a major story:
https://www.404media.co/wikipedia-pause ... -backlash/
https://archive.ph/qM9Mj

Nearly all the squawking about "Simple Article Summaries" has been on the Village Pump's technical area. Where only obsessive WP cult members will see it. SAS was announced at Wikimania last year, and was ignored. The first notice of the test I could find, from February, was totally ignored. There is a page about the SAS, which was also ignored--until this month, when notorious WikiAssholes like LilianaUwU and Nemo bis started dumping rants in the talkpage.

The real raging starts here:
https://en.wikipedia.org/wiki/Wikipedia ... bile_study
Link in case of "inevitable" archiving:
https://en.wikipedia.org/w/index.php?ti ... bile_study

One of the commentors has already been blocked as a "socik" lol, by none other than Doug Weller:
https://en.wikipedia.org/w/index.php?ti ... 1295256497

And that's not the ONLY WMF attempt to push AI natural language tools into WP editing. There's something called "Tone Check", which is also generating megabytes of arm-flapping. With the ever-lovely Tamzin leading the way.
https://en.wikipedia.org/wiki/Wikipedia ... ore_subtle

There will be more like this. Every other content generator on the web is experimenting with AI modeling, mostly to save money. Of course the "professionals" at WMF would like to be rid of its demented editor volunteer mob AND save money, so they can pay themselves Meta-level salaries. Do I need to point to the many threads on this very forum showing what results when random human beings (and their custom bots) generate content? Or the thousands of examples of "good wikipedians" abusing their authority to control and silence others? No wonder the WMF wants to ditch many of them. LLM AIs sometimes generate rants and drivel, but then, so do people. And it remains to be proven that an AI WP administrator would go on a berserk block-revert spree for the "lulz".

You can expect open warfare between the WMF employees and the "vaunted and glorious community" over AI tools.

And very soon, and it will be brutal.

The truth about AI (at least the LLM version of it) is.....it's a giant waste of time. Ed Zitron has gone to excruciating detail in multiple
articles explaining how ChatGPT, OpenAI and the rest are an endless money-sink producing products NOBODY (but scammers and lazy high school or college students) WANTS. The model is doomed to failure, and it might drag Silly Valley down with it. This is as good as a Large Language Model will get, because it needs at least two or three Internets' worth of original material to improve markedly. It cannot truly create, it has no consciousness, no sense of being and existing in the world; it is a detached simulacrum of a human mind. It shows how limited our knowledge of how the human mind works truly is because we can't fake one worth a damn.

Strelnikov wrote: ↑

Thu Jun 12, 2025 11:11 pm

The truth about AI (at least the LLM version of it) is.....it's a giant waste of time. Ed Zitron has gone to excruciating detail in multiple
articles explaining how ChatGPT, OpenAI and the rest are an endless money-sink producing products NOBODY (but scammers and lazy high school or college students) WANTS. The model is doomed to failure, and it might drag Silly Valley down with it. This is as good as a Large Language Model will get, because it needs at least two or three Internets' worth of original material to improve markedly. It cannot truly create, it has no consciousness, no sense of being and existing in the world; it is a detached simulacrum of a human mind. It shows how limited our knowledge of how the human mind works truly is because we can't fake one worth a damn.

The LLM version is used extensively in the non-AI software industry, including by professionals. I'd go so far as to say its been widely embraced there. Sometimes if only just to 'proofread' a final, entirely human created output. I mean why not, it's usually free. And when (not if but when) it finds a mistake you missed, that can only be a good thing. As long as you know enough to ignore harmful suggestions or false positives, of which there will be plenty.

Additionally, it seems near half the small businesses I email filter their responses through some sort of AI to maintain professionalism or save time. Win win for them as well. People like David Gerard are treating LLMs like the next tulip mania, a "pivot" from crypto among greedy tech bros. While that does happen, the fact that almost half the web finds a use for it on a regular basis shows it is way more useful for most people than crypto.

The main issue isn't it's lack of soul imho, although that is an issue. The main issue imho is that it takes absurd amounts of money to sufficiently train a general purpose model with a custom dataset, which should be democratized, but really isn't.

Wikimedia CAN afford to train its own model though and boot its human google aggregators ("Wikipedia editors"), could be something to do with their money they have nothing to do with.

If Wikipedia even exists 30 years from now in a remotely similar fashion, I'm convinced they'll be encouraging their users to filter their writing through an LLM to improve their output, if they don't automatically do it for them. A lot of workplaces now mandate at least proofreading LLM AIs on a browser extension among employees.

journo wrote: ↑

Fri Jun 13, 2025 1:12 am

The LLM version is used extensively in the non-AI software industry, including by professionals. I'd go so far as to say its been widely embraced there. Sometimes if only just to 'proofread' a final, entirely human created output. I mean why not, it's usually free. And when (not if but when) it finds a mistake you missed, that can only be a good thing. As long as you know enough to ignore harmful suggestions or false positives, of which there will be plenty.

Additionally, it seems near half the small businesses I email filter their responses through some sort of AI to maintain professionalism or save time. Win win for them as well. People like David Gerard are treating LLMs like the next tulip mania, a "pivot" from crypto among greedy tech bros. While that does happen, the fact that almost half the web finds a use for it on a regular basis shows it is way more useful for most people than crypto.

Yes, as long as the services are free or relatively cheap. Only a matter of time before AI operators start demanding a paid subsciption. Then 95% of their userbase will vanish. Not so long ago you could buy an Adobe application and just use it, for years; now they demand a monthly sub for online "access" to the "Adobe Creative Suite". AutoCad is even worse. So is Intuit.

The main issue isn't it's lack of soul imho, although that is an issue. The main issue imho is that it takes absurd amounts of money to sufficiently train a general purpose model with a custom dataset, which should be democratized, but really isn't.

The scraping of online data was "democratized" because the AI companies just went out and did it. Without seeking any permission or offering payment, usually. Capitalize the gains, socialize the losses. Parasitize anyone you can. Break stuff.

If Wikipedia even exists 30 years from now in a remotely similar fashion, I'm convinced they'll be encouraging their users to filter their writing through an LLM to improve their output, if they don't automatically do it for them. A lot of workplaces now mandate at least proofreading LLM AIs on a browser extension among employees.

Or they will just dump everyone and have AIs write the content. Proofreading is one thing (god knows a lot of "working journalists" need to be forcibly proofread) but making all the text and pictures is what AI companies are working toward.

journo wrote: ↑

Fri Jun 13, 2025 1:12 am

Strelnikov wrote: ↑

Thu Jun 12, 2025 11:11 pm

The truth about AI (at least the LLM version of it) is.....it's a giant waste of time. Ed Zitron has gone to excruciating detail in multiple
articles explaining how ChatGPT, OpenAI and the rest are an endless money-sink producing products NOBODY (but scammers and lazy high school or college students) WANTS. The model is doomed to failure, and it might drag Silly Valley down with it. This is as good as a Large Language Model will get, because it needs at least two or three Internets' worth of original material to improve markedly. It cannot truly create, it has no consciousness, no sense of being and existing in the world; it is a detached simulacrum of a human mind. It shows how limited our knowledge of how the human mind works truly is because we can't fake one worth a damn.

The LLM version is used extensively in the non-AI software industry, including by professionals. I'd go so far as to say its been widely embraced there. Sometimes if only just to 'proofread' a final, entirely human created output. I mean why not, it's usually free. And when (not if but when) it finds a mistake you missed, that can only be a good thing. As long as you know enough to ignore harmful suggestions or false positives, of which there will be plenty.

Additionally, it seems near half the small businesses I email filter their responses through some sort of AI to maintain professionalism or save time. Win win for them as well. People like David Gerard are treating LLMs like the next tulip mania, a "pivot" from crypto among greedy tech bros. While that does happen, the fact that almost half the web finds a use for it on a regular basis shows it is way more useful for most people than crypto.

The main issue isn't it's lack of soul imho, although that is an issue. The main issue imho is that it takes absurd amounts of money to sufficiently train a general purpose model with a custom dataset, which should be democratized, but really isn't.

Wikimedia CAN afford to train its own model though and boot its human google aggregators ("Wikipedia editors"), could be something to do with their money they have nothing to do with.

If Wikipedia even exists 30 years from now in a remotely similar fashion, I'm convinced they'll be encouraging their users to filter their writing through an LLM to improve their output, if they don't automatically do it for them. A lot of workplaces now mandate at least proofreading LLM AIs on a browser extension among employees.

You didn't read the links in my post; Zitron goes to great lengths to explain that AI is a replacement for the failed NFT experiment and the general cryptocurrency scam. The amounts of money these projects are absorbing are enormous and it will all be for products NOBODY wants. It will probably sink Silicon Valley. The whole thing is unworkable crap.

Strelnikov wrote: ↑

Fri Jun 13, 2025 4:59 pm

You didn't read the links in my post; Zitron goes to great lengths to explain that AI is a replacement for the failed NFT experiment and the general cryptocurrency scam. The amounts of money these projects are absorbing are enormous and it will all be for products NOBODY wants. It will probably sink Silicon Valley. The whole thing is unworkable crap.

I'll read the blog you linked and may comment on it later. The premise contradicts my lived experienced. I've been working in AI almost 10 years before developing any NFT software. Meaning, one came before and existed during the other. Even the worst people in the AI space, who I've been vocally critical of, have a genuine interest in the technology and aren't just grifters or profiteers.

While I wouldn't invest more than $20 in it, NFTs can be a fun way to create collaborative, immutable art. Entire virtual worlds would have been saved if built on a distributed ledger rather than some dude's homes server. A common criticism is that they are just links to art, no you can embed the art pixel data in the transcactions themselves, I posted that on Gerard's blog and he deleted it. AI encompasses a very large amount of technology, and I don't know what you mean by no one wanting LLM products, when they are being paid for extensively by businesses and consumers.

journo wrote: ↑

Fri Jun 13, 2025 5:51 pm

A common criticism is that they are just links to art, no you can embed the art pixel data in the transcactions themselves, I posted that on Gerard's blog and he deleted it.

And this "surprises" you?